A Survey of Long-term Object Tracking Methods

Visual object tracking is widely used in all walks of life. However, at present, many trackers cannot perform long-term tracking, and will lose the target in the face of occlusion, disappearance and other situations. Long term tracking can deal with this situation, so as to adapt to the complex real environment. In the field of object tracking, the current very popular tracking methods include correlation filtering method and deep learning method. This paper first introduces the basic principles of these two tracking methods, analyzes their performance in long-term object tracking, and then classifies the trackers according to non-deep learning class and deep learning class. This class of long-term tracking algorithms are described respectively. Combined with the recent popular tracking algorithms, the structure is analyzed. This paper explains its implementation mode in long-term tracking, and introduces the popular long-term tracking data sets in recent years. Finally, the current status of long-term object tracking is analyzed, and the future development direction is summarized and prospected.